Good vs. Bad Carbohydrates: How I Used a Product Analytics Approach to Tell The Difference - Part 1

Can you learn about KPIs and Metrics while eating food? How about KPIs from food labels?

What do a nutritionist, a food scientist, and a Product Manager have in common? They all make data-driven decisions.

How can you apply Product Analytics and Product Management best practices to make data-driven decisions about your food or diet? This article will help you understand KPIs, data analysis, and synthesis, all using food items - no tech, no APIs, no “Software is eating the world".

Even better, how can you avoid getting into the weeds, yet know the optimal food choices? You can use this article as a cheat sheet for choosing carbohydrate sources when on a diet.

An updated version of this with a customGPT is here.

Thumbnail credits to Wood photo created by jcomp.

Why Apply Product Analytics To Fitness?

It has been a while since I posted about applying product analytics to fitness. The process helped me learn and illustrate that we can determine a goal, define success metrics towards it, and quantify varied items to make them comparable. We literally compared apples vs oranges (oranges win!).

As it did for me, I hope going through this exercise helps you understand KPIs, learn data analysis and synthesis, and use it as a cheat sheet for choosing carbohydrate sources if/when you are maintaining a diet.

In previous articles, we’ve ranked fruits, vegetables, and even protein sources. Now we will apply a similar principle to ranking carbohydrate food sources. Carbs have been especially hard to compare because of the conflicting goals of a carbohydrate source. It took me almost two years to complete this analysis, so I will publish it now and add finality on it.

Although there is a lot of information online, The challenge in using existing sources of information online is:

1. They do not contain all the food items of interest, e.g. middle-eastern or African grains may not be included

2. They do not normalize the content, e.g. the website might compare “1 bowl” of potatoes, one tortilla, and 100g of oats

3. They do not contain pricing information

4. They do not have the tooling to compare food items

Science of Carbohydrates

We’ve covered some of the below when ranking fruits and vegetables. Food items can have three types of carbohydrates (Cleveland Clinic):

1. starch,

2. sugar, and

3. fiber.

More details on each:

1. Starch: Starch is a complex carbohydrate. It takes your body longer to break down complex carbohydrates. As a result, blood sugar levels remain stable and fullness lasts longer.

2. Sugar: Sugar is a simple carbohydrate. Your body breaks down simple carbohydrates quickly. As a result, blood sugar levels rise - and then drop - quickly.

3. Fiber: Fiber is a complex carbohydrate. Your body can’t break down fiber. Most of it passes through the intestines, stimulating and aiding digestion. Fiber also regulates blood sugar, lowers cholesterol, and keeps you feeling full longer.

Glycemic Index and Load

The other topic we should discuss before analytics is the concept of the Glycemic Index. We’d discussed the benefits of low GI in an earlier article.

As per Patrick J. Skerrett at Harvard Medical School, Glycemic Index (GI) rates carbohydrate-containing foods by how much they boost blood sugar (blood glucose). This is a metric of the rate at which eating the food item spikes blood glucose. Illustrative graph from Harvard Medical School.

However, we can’t rely on the glycemic index alone for choosing a healthy diet. Some foods, like carrot and watermelon, have a high glycemic index, but a serving contains so little carbohydrate that the effect on blood sugar is small. As per a blog by Mayo Clinic, we can use Glycemic Load (GL). GL is a numerical value that indicates the change in blood glucose levels when you eat a food item. Glycemic load can be calculated as

Glycemic load = (Carbohydrate content (g) in 100g of the food item) * (GI of the food item) / (100g)

We will review how GI and GL change with food items and other metrics we use so that it does not overlap, as we want to combine independent metrics.

Carbohydrate Goals

Let’s look at the goals for carbohydrate consumption:

1. Delay feelings of hunger

2. Energy is absorbed better into muscles

Let’s consider some anti-goals:

1. Consuming calories over one’s calorie target for the day

Tertiary objectives:

1. Spend less money for reaching the goals and against the anti-goals

2. Spend less time on the same

We will ignore the second tertiary objective. It has a lot of variability so we cannot create an objective metric for it.

How would we measure these goals?

We can look at different food items of interest and measure their:

1. Carbohydrate content

2. Calorie content

3. Glycemic Index (GI)

4. Glycemic Load (GL)

5. Expense

Which food items to consider?

I considered food items in these categories:

1. Cereal

2. Fruit

3. Grain

4. Lentil

5. Vegetable

6. Other

Across these categories, I considered about 20 food items:

Sugar, starch, and fiber distribution

The same can also be visualized as the percentage of calories that come from each. The total will not add up to 100% since protein and fat also contribute to total calories. Most of the percentage of calorie graphs show that the data is not perfect since the totals are often more than 100% for example, Banana in the below graph.

The more carbohydrate energy comes from starch and fiber, the better.

Macronutrient distribution

We’ve seen the distribution within carbohydrates, now let us see the distribution across macronutrients. The below graph shows the percentage of calories that come from different macronutrients - carbohydrate, fat, and protein.

Let’s update the graph to look proportional to the total calories. We use calories as the Y axis. Although I arranged them in descending calorie order, you’ll find discrepancies due to imperfect data.

Carbs, calories, GI, and GL

Let us review the distribution of the metrics we’ve defined above across the food items.

In general, the more carbohydrates in a food item, the higher the calories in it. However, foods higher in fat and lower in water will have higher calories for similar grams of carbohydrates, for example, Froot Loop vs White sugar. Next, let’s see a distribution of calories and GI.

Potato is to the left because of its high GI, but given the low total calories in it, I assume it will have a lower GL. Fruits and vegetables we had ranked low earlier do not seem calorie-rich, compared to grains, lentils, or cereals. We had discussed preferring GL over GI, so let us visualize GL.

The trend for GL and GI is to reduce from left to right. But, there are some interesting variations. Potato, Sweetcorn, and Banana have low GL due to the low calories. Corn Flakes is very high in GL and GI, surprisingly more than white sugar. I searched across a few websites and that held. We also see cereals to the left and lentils to the right, so maybe there is a distribution of categories?

We see that the cereals are more towards the left with higher GL whereas lentils, vegetables, and fruits are towards the right with lower GL.

Unlike vegetables or fruits, grains and lentils are usually cooked with a lot of water, which reduces the calories per gram of cooked product. I’ve seen them being cooked with a 1:2 or 1:3 ratio of item:water. So if the GL reduces to one-third or one-fourth of the number here, respectively, 1) lentils would be significantly below these high-calorie vegetables and 2) fruits and grains would be about equal. However, we will not use measurements for grains and lentils for cooked food items since numbers available online may not be your way of cooking them.

Expense

Let’s also look at the expense distribution of the food items as for the same benefits, we can prefer a food item with the lower expense. I’ve used expense and price interchangeably here.

What’s Next?

When looking at customer experience and product analytics, I find it beneficial to review the distribution of different metrics before defining the success thresholds or targeting a group of customers. Now, we’ve reviewed the distribution of metrics, we will review what success looks like and rank the carb sources in a follow-up article.